聖心女子大学国際交流学科
2024年秋学期
祝: ノーベル経済学賞受賞3人(2019年)
受賞理由:
「貧困緩和に実験的手法を導入した功績。貧困(という大きな)問題を小さな扱いやすい問題に分解し、実験を使って対策を示した。」
長所
貢献
短所
短所
倫理上、母乳育児や母乳育児に金銭的誘因を与える「推奨」を実験できないが、母乳育児の非金銭的「推奨」(内容伝達)を実験しても倫理的に問題ない
ただし、推奨内容が統御群(比較対象)の女性に伝わらないか統御は難しい
実験がうまくいき、効果があると分かっても、政策に採用されるかは別問題
政治家および投票基盤が政策実施=得策と思わねばならないから
実験は思いついた政策に効果があるか気軽に試せるが、必要となる作業監理と予算は多いために、本当に検討する価値のある政策を選ばないと資源の無駄
投与と反応を見る疫学研究ではなく、人々の意志決定と行動選択を含む経済学研究なので計測に費用がかかる
この記事には問題があります。何でしょうか。
問題を理解するためには、インパクト評価のキー・タームを知る必要があります
\(A\)と\(B\)が相関: さまざまな因果関係があり得る A correlation between \(A\) and \(B\) can include multitude of cases.
笑いは健康(自己申告)を引き起こすか?
Does laughter cause (self-reported) healthiness?
この研究のデザインは因果関係を明らかにできるか?
因果関係と相関関係は同じではない
\[ \begin{aligned} \mbox{試験点数}&=20+20*D+e D&= \left\{ \begin{array}{c} 0 1 \end{array} \right. \quad \mbox{if 塾に週1時間以上} \left\{ \begin{array}{l} \mbox{通わない} \mbox{通う} \end{array} \right. \end{aligned} \]
\(D\)はダミー変数dummy variableと呼ばれる0と1の2つの値をとる離散変数。2つの値しかとらないので2項変数binary variableとも呼ばれる。ここでは塾に週1時間以上通うと1、そうではない場合は0という値をとる変数。塾に通う人と通わない人にグループ分けできる。\(e\)は誤差を表す確率変数で誤差項error termという。
ダミー変数は質的情報を表現できる:
\[ \mbox{試験点数}=20+20*D+e \]
これが因果関係の場合: 塾に通う\(\Rightarrow\)点数が40点になる、という解釈になる。
でも、方程式は相関関係を表す場合もある。
「勉強好き」という欠落変数omitted variableが試験点数と塾通いに同時に影響を与えていて、試験点数と塾通いの間に因果的な関係はない。
方程式は逆の因果関係を表す場合もある。
この場合、点数\(\Rightarrow\)塾通いという逆の因果関係が成り立っている。
方程式は必ずしも右辺\(\Rightarrow\)左辺の因果関係ばかりではなく、逆方向の因果関係や欠落変数を通じた相関関係も含む。
因果関係を示すためには特定の条件が必要。その条件がない通常の回帰式の場合、相関関係までしか読み取ることができない。
予測だけなら相関関係で十分
でも、相関関係からは理由やメカニズム(因果関係のどの組み合わせか)は分からない
相関関係に頼った予測はメカニズムを検討しないため、理論なき計測measurement without a theoryと揶揄されることもある
因果関係を示す方法: ランダム化統御試験randomised controlled trial (RCT)
プロジェクト評価の報告書にはさらっとこんな結論が散見される
「プロジェクトによって健康状態が改善された」“with the project, health status improved.”
インパクトを知る上で適切な比較か? Is this a legit comparison?
殆どの場合、不適切 Almost always, no. → 自己選抜の図
下記のようにインパクトを計測できれば、政策が結果指標\(y_{i}\)を変えたといえる
We can say a policy changed \(y_{i}\) when we can compute its causal impact on \(y_{i}\) as
(\(i\)が政策に影響されたときの\(y_{i}\))\(-(i\)が政策に影響されなかったときの\(y_{i}\))
\[ \begin{aligned} (y_{i}|D_{i}=1) &- (y_{i}|D_{i}=0), \quad \mbox{or,} y_{i1}&-y_{i0}.%, \quad y_{i0}=y_{i}|D=0, \ y_{i1}=y_{i}|D=1. \end{aligned} \]
個人\(i\)の治療効果treatment effect of policy for individual \(i\):
\[ (y_{i}|D_{i}=1)-(y_{i}|D_{i}=0)= y_{1i}-y_{0i}. \]
プログラム評価での根源的問題 The fundamental problem in program evaluation
治療群に属するときの\(y_{i}\)と統御群に属するときの\(y_{i}\)を同時に観察できない We cannot observe \(y_{i}\) in the treated and in the control for the same individual \(i\) simultaneously.
\(\Leftrightarrow\)
各個人\(i\)の結果指標\(y_{i}\)の
\(\Leftrightarrow\)
言い換えれば、仮定なしには政策の効果を計算することはできないIn other words, we cannot compute the causal impacts of a policy for each individual \(i\) without further assumptions.
CFは何か? What are the CFs?
何も仮定しないと、個人\(i\)の政策効果は計算できない
しかし、政策裨益をランダム化をすると、政策の平均治療効果average treatment effect (ATE)は推計できるBut under treatment randomisation, we can estimate the average causal impacts of a policy, the average treatment effect (ATE). \[ ATE=\E[y_{i}|D_{i}=1]-\E[y_{i}|D_{i}=0]. \]
個人\(i\)のではなく、個人\(i\)の属する母集団の平均的な治療効果
\(\E\)は母集団全体で平均を取っていることを示す期待値記号\(\E\) is an expectation operator that indcates we are taking the mean over the entire population that \(i\) belongs to
以下の思考実験を考えるConsider the following thought experiment.
ATEを(一致推計量consistent estimator [標本サイズが無限大になると真の値になる推計量]として)得る条件 Conditions that make the ATE estimate consistent are
2.は単純化のために利用。グループごとにインパクトが違うなら、グループをもっと細かく分ければいい。2. is used for simplification. If the impact is different across subgroups, we can use finer grouping.
1.が最も重要。ランダムに割り振ることによって、各グループの特徴の分布が近似。1. is of most importance at this stage. It is randomisation of treatment status among individuals that gives similarity in distributions.
実験をしてからの手順
実験をしてからの手順
統計的推論で得る\(p\)値(\(p\) value)は帰無仮説が成り立つ確率と考えていい
5%をカットオフ
と表現されることが多いが、推奨できない
\(\leftarrow\) 4.99%と5.01%の差は無視可能なのに表現が違いすぎる、四捨五入するとき4.4%と4.5%の差、もしくは、4.95%と4.94%の差
実験をしてからの手順
ランダム化確認: permutation test (並べ替え検定), randomisation test (確率化検定)
帰無仮説null hypothesis: グループaの分布=グループbの分布
実験をしてからの手順
ランダム化確認: permutation test (並べ替え検定), randomisation test (確率化検定)
バングラデシュ最貧困層への貸付実験: 大規模貸付グループと小規模貸付グループの比較
>p1cm<>p.25cm<>p10cm< Source:& & Estimated with GUK administrative and survey data. Notes: & 1. & R’s package coin is used for baseline group mean covariates to conduct approximate permutation tests. &2. & Number of repetition is set to 100000. Step-down method is used to adjust for multiple testing of a multi-factor grouping variable. 40 are lost to flood before arm assignment.
実験をしてからの手順
他要因排除の確認 (placebo tests, falsification tests)
治療対象選定をランダム化するとATEの一致推計量が得られるRandomisation of treatment status will give us a consistent ATE estimate
われわれが計測できるのは非同意者を含むグループ平均値の差。非同意者がいるとインパクトが小さくなる。What we can measure is the mean group difference inclusive of noncompliance. Noncompliance makes estimated impacts smaller.
非同意者を含む効果推計値を治療意図に基づく効果intention-to-treat (ITT) effectという。The estimator under partial compliance is called intention-to-treat (ITT) effect, and is like a down-to-earth version of ATE.
さまざまな効果推計量(実証研究の大半がITTかLATE)
\[ \mbox{割り当ての結果への効果}=\mbox{割り当ての参加への効果}\times\mbox{参加の結果への効果} \]
\[ \begin{aligned} \mbox{参加}\Rightarrow\mbox{結果} &= \frac{\mbox{割り当て}\Rightarrow\mbox{結果}}{\mbox{割り当て}\Rightarrow\mbox{参加}}, LATE=\mbox{介入}\Rightarrow\mbox{結果} &= \frac{\mbox{割り当て}\Rightarrow\mbox{結果}}{\mbox{割り当て}\Rightarrow\mbox{介入}}, &= \frac{\mbox{割り当てによる処置群と統御群の結果の差}\alpha\mu - \beta\mu}{\mbox{割り当てによる処置群と統御群の介入比率の差}\alpha-\beta}, &=\frac{(\alpha - \beta)\mu}{\alpha-\beta}=\mu \end{aligned} \]
実験には人が意図して実施する科学実験と偶然発生する自然実験natural experimentsがある
科学実験は非倫理的なものは実施しない
自然実験は意図せず発生するので非倫理的であっても実施されてしまう
非倫理的な自然実験の例:
親の違いによる子の純資産額、所得、学歴、金融投資への影響
Fagereng, Mogstad, and Rønning (2021): 韓国の生活苦の乳児が養子縁組でノルウェイに行く
NGO: ノルウェイで養親候補を書類審査+面接、合格者の書類を韓国に送付、先着順で子どもと縁組
養親は子どもに関する希望を出せず、到着順はランダム
養親は年齢、学歴、所得、純資産額などで異なる
ランダムであることの確認:
養子全員がtreatedなのでpermutation testは使えない
親の違いによる子の純資産額、所得、学歴、金融投資への影響
親の違いによる子の純資産額、所得、学歴、金融投資への影響
養子\(i\)の特徴\(Y_{i}\)と養親\(j\)の\(k+1\)個の特徴\(W_{j}, x_{1j}, \dots, x_{kj}\)がどのように関係しているかを見たい (養親純資産額は\(W_{j}\), 養子の特徴も\(m\)個: \(x_{1i}, \dots, x_{mi}\))
親の違いによる子の純資産額、所得、学歴、金融投資への影響
\[ \begin{aligned} Y_{i} &= \overbrace{\alpha_{1965}Z_{1965}+\cdots+\alpha_{1986}Z_{1986}}^{\scriptsize{\mbox{縁組年の固定効果}}}+{\color{red}\beta} W_{j} &\hspace{1em} +\underbrace{\eta_{1}x_{1j}+\cdots+\eta_{k}x_{kj}}_{\scriptsize{\mbox{養親の特徴の効果}}} &\hspace{1em} +\underbrace{\lambda_{1}x_{1i}+\cdots+\lambda_{m}x_{mi}}_{\scriptsize{\mbox{養子の特徴の効果}}} \\[-3ex] &\hspace{1em} +\underbrace{\gamma\kappa_{j}+\delta\chi_{i}}_{\scriptsize{\mbox{養親と養子の各個人固定効果}}}+u_{i}. \end{aligned} \]
\({\color{red}\beta}\): 養親から養子への純資産額^の伝播係数
\(\beta\)は以下の効果を除いた上で推計されたものです
親の違いによる子の純資産額、所得、学歴、金融投資への影響
OLS推計結果
親の違いによる子の純資産額、所得、学歴、金融投資への影響
所得や学歴以外の家庭内の何かが子の純資産額を高める
参考: Barth, Papageorge, and Thom (2020) take aways
株式所有経由で資産格差を説明するのは共通
グループの割り振り(\(D_{i}\))がランダム化していないと、ごく稀なケースを除き、政策がない場合の結果指標の分布はグループ間で異なる。When the treatment assignment (\(D_{i}\)) is not randomised, except for very rare lucky cases, the distributions of outcome measure in the absense of a policy are different between the treated and the control.
被験者=目的意識を持って参加する人間なので、参加者と不参加者は特徴が異なるThis is because we are dealing with humans who participate purposefully.
What we will learn:
Implementation:
Get data:
Stjpg:
A simpler way: \(y^{DID}=a_{3}\) of the following regression.
回帰式って何?
説明するよりも見る方が分かりやすいので、まずはデータを用意します
wooldridgeパッケージをインストールlibrary(wooldridge)と入力しdata("wagepan")でRに読み込まれますデータの種類
package 'wooldridge' successfully unpacked and MD5 sums checked
The downloaded binary packages are in
C:\Users\acorn\AppData\Local\Temp\Rtmp2XWEwR\downloaded_packages
nr year agric black bus construc ent exper fin hisp poorhlth
<int> <int> <int> <int> <int> <int> <int> <int> <int> <int> <int>
1: 13 1980 0 0 1 0 0 1 0 0 0
2: 13 1981 0 0 0 0 0 2 0 0 0
3: 13 1982 0 0 1 0 0 3 0 0 0
4: 13 1983 0 0 1 0 0 4 0 0 0
5: 13 1984 0 0 0 0 0 5 0 0 0
---
4356: 12548 1983 0 0 0 1 0 8 0 0 0
4357: 12548 1984 0 0 0 1 0 9 0 0 0
4358: 12548 1985 0 0 0 1 0 10 0 0 0
4359: 12548 1986 0 0 0 0 0 11 0 0 0
4360: 12548 1987 0 0 0 0 0 12 0 0 0
hours manuf married min nrthcen nrtheast occ1 occ2 occ3 occ4 occ5
<int> <int> <int> <int> <int> <int> <int> <int> <int> <int> <int>
1: 2672 0 0 0 0 1 0 0 0 0 0
2: 2320 0 0 0 0 1 0 0 0 0 0
3: 2940 0 0 0 0 1 0 0 0 0 0
4: 2960 0 0 0 0 1 0 0 0 0 0
5: 3071 0 0 0 0 1 0 0 0 0 1
---
4356: 2080 0 1 0 0 0 0 0 0 0 1
4357: 2080 0 1 0 0 0 0 0 0 0 1
4358: 2080 0 1 0 0 0 0 0 0 0 1
4359: 2080 0 1 0 0 0 0 0 0 0 1
4360: 3380 0 1 0 0 0 0 0 0 0 1
occ6 occ7 occ8 occ9 per pro pub rur south educ tra trad
<int> <int> <int> <int> <int> <int> <int> <int> <int> <int> <int> <int>
1: 0 0 0 1 0 0 0 0 0 14 0 0
2: 0 0 0 1 1 0 0 0 0 14 0 0
3: 0 0 0 1 0 0 0 0 0 14 0 0
4: 0 0 0 1 0 0 0 0 0 14 0 0
5: 0 0 0 0 1 0 0 0 0 14 0 0
---
4356: 0 0 0 0 0 0 0 1 1 9 0 0
4357: 0 0 0 0 0 0 0 1 1 9 0 0
4358: 0 0 0 0 0 0 0 1 1 9 0 0
4359: 0 0 0 0 0 1 0 1 1 9 0 0
4360: 0 0 0 0 0 0 1 1 1 9 0 0
union lwage d81 d82 d83 d84 d85 d86 d87 expersq
<int> <num> <int> <int> <int> <int> <int> <int> <int> <int>
1: 0 1.197540 0 0 0 0 0 0 0 1
2: 1 1.853060 1 0 0 0 0 0 0 4
3: 0 1.344462 0 1 0 0 0 0 0 9
4: 0 1.433213 0 0 1 0 0 0 0 16
5: 0 1.568125 0 0 0 1 0 0 0 25
---
4356: 0 1.591879 0 0 1 0 0 0 0 64
4357: 1 1.212543 0 0 0 1 0 0 0 81
4358: 0 1.765962 0 0 0 0 1 0 0 100
4359: 1 1.745894 0 0 0 0 0 1 0 121
4360: 1 1.466543 0 0 0 0 0 0 1 144
The data was used for estimating union impacts on wages. 1980 to 1987.
nr, agric, black, construc, exper, hisp, poorhlth, hours, manuf, educ, union, lwage.dunion == 0, control), or union == 0 in 1980 and union == 1 in 1987 (dunion == 1, treated).keepvar <- c("nr", "year", "agric", "black", "construc", "exper", "hisp", "poorhlth",
"hours", "manuf", "educ", "union", "lwage")
wagepan <- wagepan[, keepvar, with = F]
wagepan <- wagepan[year == 1980 | year == 1987, ]
setkey(wagepan, nr, year)
wagepan[, dunion := diff(union), by = nr]
wagepan <- wagepan[dunion >= 0 & union[year == 1980] == 0, ]
table(wagepan[, dunion])
0 1
612 112
factorcols <- c("agric", "black", "construc", "hisp", "poorhlth", "manuf", "union")
wagepan[, (factorcols) := lapply(.SD, as.factor), .SDcols = factorcols]
wagepan[, after := 0L]
wagepan[year == 1987, after := 1L]
summary(wagepan) nr year agric black construc exper
Min. : 13 Min. :1980 0:696 0:636 0:668 Min. : 0.000
1st Qu.: 2540 1st Qu.:1980 1: 28 1: 88 1: 56 1st Qu.: 3.000
Median : 4718 Median :1987 Median : 8.000
Mean : 5392 Mean :1984 Mean : 6.554
3rd Qu.: 8568 3rd Qu.:1987 3rd Qu.:10.000
Max. :12548 Max. :1987 Max. :18.000
hisp poorhlth hours manuf educ union
0:613 0:710 Min. : 120 0:531 Min. : 3.00 0:552
1:111 1: 14 1st Qu.:1956 1:193 1st Qu.:11.00 1:172
Median :2080 Median :12.00
Mean :2140 Mean :11.76
3rd Qu.:2410 3rd Qu.:12.00
Max. :4992 Max. :16.00
lwage dunion after
Min. :-1.114 Min. :0.0000 Min. :0.0000
1st Qu.: 1.305 1st Qu.:0.0000 1st Qu.:0.0000
Median : 1.648 Median :0.0000 Median :1.0000
Mean : 1.622 Mean :0.1547 Mean :0.5028
3rd Qu.: 1.991 3rd Qu.:0.0000 3rd Qu.:1.0000
Max. : 3.313 Max. :1.0000 Max. :1.0000
Key: <nr, year>
Index: <year>
nr year agric black construc exper hisp poorhlth hours manuf educ
<int> <int> <int> <int> <int> <int> <int> <int> <int> <int> <int>
1: 13 1980 1 1 1 1 1 1 2672 1 14
2: 13 1987 1 1 1 8 1 1 2640 1 14
3: 17 1980 1 1 1 4 1 1 2484 1 13
4: 18 1987 1 1 1 11 1 1 2340 1 12
5: 120 1980 1 1 1 2 1 1 1025 1 10
---
720: 12500 1980 1 1 1 4 1 1 2008 1 12
721: 12500 1987 1 1 1 11 1 1 3276 2 12
722: 12534 1987 1 1 2 9 1 1 2080 1 11
723: 12548 1980 1 1 2 5 1 1 2000 1 9
724: 12548 1987 1 1 1 12 1 1 3380 1 9
union lwage dunion after
<int> <num> <int> <int>
1: 1 1.1975402 0 0
2: 1 1.6691879 0 1
3: 1 1.6759624 0 0
4: 1 2.8731608 0 1
5: 1 0.2585549 0 0
---
720: 1 0.9724026 0 0
721: 1 1.3067402 0 1
722: 1 2.3429170 0 1
723: 1 1.1305454 1 0
724: 2 1.4665428 1 1
DID estimation: union参加(union=0→1)というtreatmentが賃金に与える効果
Call:
lm(formula = lwage ~ union + after + I(union * after), data = wagepan)
Residuals:
Min 1Q Median 3Q Max
-2.44510 -0.26950 0.03016 0.32441 1.49050
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1.01422 0.08545 11.870 < 2e-16 ***
union 0.31706 0.07036 4.507 7.69e-06 ***
after 0.69315 0.11481 6.038 2.51e-09 ***
I(union * after) -0.20088 0.08927 -2.250 0.0247 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.4873 on 720 degrees of freedom
Multiple R-squared: 0.217, Adjusted R-squared: 0.2137
F-statistic: 66.5 on 3 and 720 DF, p-value: < 2.2e-16
*に別の意味、文字通りのかけ算にはI()を付けるCompute DID by hand…union wages are high at start
Key: <union, after>
union after meanw
<int> <int> <num>
1: 1 0 1.331278
2: 1 1 1.823554
3: 2 0 1.648341
4: 2 1 1.939739
Key: <union>
union dmean.lwage
<int> <num>
1: 1 0.4922759
2: 2 0.2913973
[1] -0.2008786
回帰式の利点Advantages of regression
DID estimation with covariates: \(k\)個の他要因\(x_{1i}, \dots, x_{ki}\)を制御してunion参加(union=0→1)というtreatmentが賃金に与える効果
\[ y_{i,t}=a_{0}+a_{1}D_{i}+a_{2}after_{t+1}+a_{3}D_{i}*after_{t+1}+b_{1}x_{1i}+\dots+b_{k}x_{ki}+e_{i,t} \]
did2 <- lm(data = wagepan, lwage ~ union + after + I(union*after) + educ + exper + hours + agric + construc + manuf + black + hisp + poorhlth)
summary(did2)
Call:
lm(formula = lwage ~ union + after + I(union * after) + educ +
exper + hours + agric + construc + manuf + black + hisp +
poorhlth, data = wagepan)
Residuals:
Min 1Q Median 3Q Max
-2.60937 -0.25180 0.05089 0.27899 1.43828
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -2.912e-01 3.033e-01 -0.960 0.337363
union 2.834e-01 6.761e-02 4.192 3.12e-05 ***
after 2.659e-01 1.437e-01 1.851 0.064570 .
I(union * after) -1.527e-01 8.445e-02 -1.808 0.071020 .
educ 1.012e-01 1.219e-02 8.302 5.17e-16 ***
exper 4.968e-02 1.288e-02 3.858 0.000125 ***
hours -2.462e-05 2.880e-05 -0.855 0.392900
agric -1.703e-01 9.156e-02 -1.860 0.063294 .
construc 1.123e-01 6.663e-02 1.685 0.092385 .
manuf 1.608e-01 4.055e-02 3.965 8.07e-05 ***
black -1.327e-01 5.466e-02 -2.428 0.015450 *
hisp 2.194e-02 4.914e-02 0.447 0.655344
poorhlth 4.170e-02 1.254e-01 0.333 0.739517
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.4591 on 711 degrees of freedom
Multiple R-squared: 0.3136, Adjusted R-squared: 0.302
F-statistic: 27.07 on 12 and 711 DF, p-value: < 2.2e-16
Summarise estimated results
package 'modelsummary' successfully unpacked and MD5 sums checked
The downloaded binary packages are in
C:\Users\acorn\AppData\Local\Temp\Rtmp2XWEwR\downloaded_packages
package 'tinytable' successfully unpacked and MD5 sums checked
The downloaded binary packages are in
C:\Users\acorn\AppData\Local\Temp\Rtmp2XWEwR\downloaded_packages
| simple | with covariates | |
|---|---|---|
| (Intercept) | 1.014 | -0.291 |
| (<0.001) | (0.337) | |
| union | 0.317 | 0.283 |
| (<0.001) | (<0.001) | |
| after | 0.693 | 0.266 |
| (<0.001) | (0.065) | |
| I(union * after) | -0.201 | -0.153 |
| (0.025) | (0.071) | |
| educ | 0.101 | |
| (<0.001) | ||
| exper | 0.050 | |
| (<0.001) | ||
| hours | 0.000 | |
| (0.393) | ||
| agric | -0.170 | |
| (0.063) | ||
| construc | 0.112 | |
| (0.092) | ||
| manuf | 0.161 | |
| (<0.001) | ||
| black | -0.133 | |
| (0.015) | ||
| hisp | 0.022 | |
| (0.655) | ||
| poorhlth | 0.042 | |
| (0.740) | ||
| Num.Obs. | 724 | 724 |
| R2 | 0.217 | 0.314 |
| R2 Adj. | 0.214 | 0.302 |
| AIC | 1019.7 | 942.3 |
| BIC | 1042.6 | 1006.5 |
| Log.Lik. | -504.854 | -457.163 |
| F | 66.499 | 27.072 |
| RMSE | 0.49 | 0.45 |
DIDの識別仮定: 介入がないとき、結果指標の変化が群間で似ていること。共通トレンドの仮定 common trend assumption.
識別仮定List of identifying assumptions.
効果を識別identifyしているならば成立しなくてはならない仮定: どのくらい現実的か=推計値の信頼度credibility of estimate
In the absence of the policy: \(\delta(A)\) reads “distribution of \(A\)”
For DID we need:
You need to get the control data even before the policy starts.
Before-after:
With-without:
With-without:
Before-after: